DMCA
Time series feature extraction for data mining using DWT and DFT (2003)
| Citations: | 28 - 1 self |
BibTeX
@TECHREPORT{Mörchen03timeseries,
author = {Fabian Mörchen},
title = {Time series feature extraction for data mining using DWT and DFT},
institution = {},
year = {2003}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
A new method of dimensionality reduction for time series data mining is proposed. Each time series is compressed with wavelet or Fourier decomposition. Instead of using only the first coefficients, a new method of choosing the best coefficients for a set of time series is presented. A criterion function is evaluated using all values of a coefficient position to determine a good set of coefficients. The optimal criterion function with respect to energy preservation is given. For many real life data sets much more energy can be preserved, which is advantageous for data mining tasks. All time series to be mined, or at least a representative subset, need to be available a priori.
Keyphrases
time series feature extraction time series new method fourier decomposition criterion function energy preservation data mining task coefficient position dimensionality reduction optimal criterion function time series data mining representative subset first coefficient many real life data set good set
